Experiment Review
Use this when
A campaign or growth test has reached its decision date or produced enough evidence for a review.
Process
- Restate the original hypothesis, primary metric, baseline, threshold, audience, and time window.
- Check tracking quality, sample composition, external changes, and missing data before interpreting results.
- Compare the primary outcome with the threshold and use supporting metrics to diagnose why.
- Separate observations from explanations and state the confidence of each explanation.
- Decide to stop, iterate, scale, or gather more data.
- Define the next action, owner, cost cap, and what would change the decision.
Output
Include decision, evidence, data-quality notes, interpretation, confidence, lessons, and the next experiment or rollout step.
Guardrails
Do not cherry-pick supporting metrics, hide negative outcomes, or imply causation from correlation. Avoid false precision when sample size or tracking is weak.
1---2name: experiment-review3description: Review a growth experiment and make an evidence-based stop, iterate, or scale decision.4---56# Experiment Review78## Use this when910A campaign or growth test has reached its decision date or produced enough evidence for a review.1112## Process13141. Restate the original hypothesis, primary metric, baseline, threshold, audience, and time window.152. Check tracking quality, sample composition, external changes, and missing data before interpreting results.163. Compare the primary outcome with the threshold and use supporting metrics to diagnose why.174. Separate observations from explanations and state the confidence of each explanation.185. Decide to stop, iterate, scale, or gather more data.196. Define the next action, owner, cost cap, and what would change the decision.2021## Output2223Include decision, evidence, data-quality notes, interpretation, confidence, lessons, and the next experiment or rollout step.2425## Guardrails2627Do not cherry-pick supporting metrics, hide negative outcomes, or imply causation from correlation. Avoid false precision when sample size or tracking is weak.
milind-soni